History of Digital Game AI (2)(Intelligent Man Machine Interaction)

History of Digital Game AI (2)(Intelligent Man Machine Interaction)

Continuing on from my previous article on digital game AI, this time I would like to discuss the autonomous agents that appeared in the last generation (after 2000).

Textbook of Digital Games: The Latest Trends in the Game Industry You Need to Know

In the last generation (after 2000), there was a lot of architecturalization of AI. First of all, the “Agent” architecture was gradually formed from the 80s to the 90s. This refers to an AI that is given a role to fulfill a purpose, and is synonymous with a character in the game world.

In this context, an autonomous agent is one that has the ability to elaborate on its own goals while assessing the surrounding environment and situation. These agents are called Multi-Agent.

The C4 architecture of MIT is one of the most famous examples of these agent technologies. The C4 architecture is an architecture proposed by the Synthetic Characters Group of MIT in 2001 to realize the intelligence of virtual animals in virtual space. As shown below, each module (knowledge source, KS) is connected to the blackboard architecture.

The goal of this system is to create a pet-type agent that learns, interacts with humans, and has independence. The purpose of this project is to explore the architecture of the system. This framework was applied in the game world as the basic framework for FPS, and became a trend throughout the 2000s.

The basic idea of the agent architecture is to first clearly distinguish between the outside world (the game world) and the AI, and then define sensors (receptors) that have the function of collecting information from the outside world and effectors (effectors) that influence the outside world.

Next, based on the information obtained from the sensors, the AI constructs (models) the cognitive world, thinks and makes decisions based on this model, designs actions based on these decisions, and outputs action commands (or action sequences). These are finally communicated to the effectors to become actions.

In the case of C4, the action commands are once written on the blackboard, and the effector (body) reads them.

With such a modular structure, for example, various algorithms such as conditional reflexes, planning techniques, and finite state machines can be applied to the thinking part.

When this is applied to a FPS, a hierarchical FSM (HFSM) and goal-oriented action planning are introduced to enable fast situational judgment and action in order to cope with combat with gunfire. (Notable examples include the HALO series, F.E.A.R., Killzone, etc.)

As an example of application other than FPS, “The Sims” series has also been applied to agents that live their daily lives. This is a game in which the AI lives in a box-like world consisting of multiple houses, and the user can enjoy the interaction between the AI with occasional interference, and unlike FPS, it has the different difficulty of having the intelligence to spend the slow time of daily life as naturally and interestingly as possible.

In particular, Ai differs from FPS in that it is characterized by its internal richness. Unlike the extreme state of “defeating the enemy” in FPS, Ai has desires such as “I want to talk to people”, “I am lonely”, “I am hungry”, “I want to clean my room”, etc., and while these desires change over time, there are various actions such as “talking to someone”, “watching TV”, “cleaning”, “sleeping”, etc.

These everyday agents are formed by carefully creating the relationship between the object, the agent’s actions on the object, and the emotions satisfied by the actions. In “The Sims”, each object is accompanied by a list of “actions” that the agent takes with the object and “changes in emotions when the actions are taken”, and the agent selects the object that maximizes the total score (parameter of desire fulfillment) calculated from the agent’s current state. The agent selects the object that increases the total point (parameter of desire satisfaction) calculated from its current state the most and acts accordingly.

With this kind of mechanism, the agent will move around autonomously according to the changes in its internal state.